This document presents initial results for a health-impact assessment related to adding more vegetated landscape to the City of Denver. We aim to estimate the number of deaths that would be prevented by adding more green space, specifically native plants, to the City of Denver. We estimate the number of deaths prevented corresponding to four categories of greening scenarios, which were informed by conversations with local stakeholders. The scenarios are:
Add native plants to all census block groups such that
Create native buffers around riparian areas (rivers, streams, lakes, and ponds) of the following sizes:
Initiatives related to green infrastructure and stormwater management. (Please refer to corresponding section for details.)
Replace a portion of parking-lot surface with native plants:
For all scenarios, we measured green-space exposure by the normalized difference vegetation index (NDVI) measured by the Landsat-8 satellite at a spatial resolution of 30 meters squared. To determine the target (alternative) NDVI value, we measured the NDVI of a 100% native zone in the Denver Botanic Gardens on several cloud-free spring and summer days. The NDVI value was about 0.5 in this area, so we assumed that, by adding more native plants to places in Denver, the NDVI of those areas would change to 0.5. Other less maintained areas may have a lower NDVI, so we plan to vary
Under the first scenario, which we view as the most ambitious, we consider the expected impact on mortality by applying a green-space intervention to all census block groups, without specifying where. This scenario was informed by conversations with colleagues who stated that
We first measured the mean NDVI of each census block group on July 4, 2021. The weighted block-group-level mean is presented here, weighted by the proportion each 30 square-meter pixel covers by the census block group. For example, if half of a pixel overlaps the block group, it receives a weight of 0.5 in the weighted average. We also removed bodies of water before measuring NDVI.
We excluded census block whose baseline NDVI was above the native-plant threshold:
We measured NDVI in a 200-foot buffer, a 100-foot buffer, and a 50-foot buffer around all bodies of water in Denver. We downloaded bodies of water from OpenStreetMap (code here. We defined residential exposure to these riparian areas as those individuals living within a 500-meter buffer, following the green-space literature that has defined green-space exposure based on residential proximity. We estimated the number of people in this buffer by multiplying the population density of the census block group by the intersecting area.
The below map depicts mean NDVI in the portions of census block groups that intersect a 200-foot buffer as well as those pieces that intersect the part of the 500 m buffer that would not be intervened upon, i.e., the part between the 200-foot buffer and the edge of the 500 m buffer.
We spoke with representatives at the local Office of Green Infrastructure, and they described three categories of initiatives–some planned, some aspirational–that could include native or adapted plants.
We were provided a list of planned projects throughout Denver. Based on our conversations, we assume that about 75% of the project’s footprint would consist of native or vegetation, so our intervention is defined as changing the NDVI values of these polygons from their baseline value (on July 4 2021) to the native-plant value of 0.5 As we did for the riparian areas, we defined exposure to the projects as those individuals living within a 500 m buffer of the projects. A map of baseline NDVI of the projects themselves and of a 500 m buffer around the projects is below.
to-do
The approach for estimating the health impact of the redevelopment controls is different from scenarios 2 and 3.a. because we do not know where the re-developments will occur. We thus simulated possible scenarios. As stated above, we anticipate the following number of parcels will be subject to these rules per year:
We gathered data on parcels from Denver’s Open Data Portal (Existing Landuse 2018) and measured their area. A subset of these parcels near Union Station is mapped below.
Then, from all parcels, we sampled 100 sites of size greater than 1 acre, 25 sites of size 0.5-1 acre, and 400 sites less than 0.5 acres. One such sample appears below.
From this point, we followed the same framework as we’ve done for other projects. We estimated that about 50% of this redevelopment would be native (subject to change), and we considered those exposed to these re-developments as anyone living within 500 m of them. The baseline NDVI of the sampled parcels and that of their corresponding 500m buffers is visualized below.
We obtained spatial data on parking lots from the City of Denver Open Data Catalog. About 9% of Denver’s area is a parking lot:
| area_mi2_prkng | den_area_mi2_total | prop_den_parking |
|---|---|---|
| 13.57 | 155 | 0.09 |
We similarly measured the NDVI on the parking lots as well as the NDVI in the census block groups that are within a 500-m buffer radius of any parking lots (which is most).
We present a small subset near Union Station, as the file size visualizing the NDVI measurements for the full city exceeds GitHub’s file-upload limits:
We estimated the number of deaths averted under each scenario by following a recent meta-analyses that estimated that for every 0.1 unit increase in exposure to NDVI, the relative risk of death decreases about 4% (pooled risk ratio of 0.96, 95% confidence interval (CI): 0.94, 0.97)(Rojas-Rueda et al. 2019). We defined NDVI exposure as that within a 500-meter buffer of the individual’s place of residence. We used this risk ratio to estimate the population attributable fraction corresponding to the proposed change in NDVI from the baseline level to the native-plants level for each census block group (first scenario) or the part of the census block group intersecting the proposed intervention area (riparian and parking scenarios). We then estimated the number of deaths prevented considering estimates of population and all-cause mortality rates. We used 5-year American Community Survey Data (2015-2019) to estimate the population in 5-year age groups in each census block group and age-stratified estimates of mortality rates for the state of Colorado from the Global Burden of Disease project. If an intervention area did not cover an entire census block group, we multiplied the population density of the block group in that age group by the area of the block group covered by the intervention area to estimate the total population in the area affected by the intervention. We restricted analyses to adults aged 30 and above following the age range of many of the cohort studies reviewed. We also restricted analyses to those census block groups or census-block-group pieces below the native-plants NDVI threshold.
| scenario | scenario_sub | pop_affected | attrib_deaths |
|---|---|---|---|
| all-bg | 20-pct | 367077.0 | -44.1 |
| all-bg | 100-pct | 367077.0 | -227.7 |
| riparian | 200-ft | 171750.2 | -7.4 |
| riparian | 100-ft | 171750.2 | -2.4 |
| riparian | 50-ft | 171750.2 | -1.0 |
| ogi | ogi_proj-pct-75 | 72584.9 | -1.1 |
| ogi | parcel-pct-50 | 300486.8 | -1.0 |
| prkng | pct-prkng-100 | 360111.6 | -40.0 |
| prkng | pct-prkng-50 | 360111.6 | -19.9 |
| prkng | pct-prkng-20 | 360111.6 | -7.9 |
Add equity component
Add economic-impact attributable to health
Incorporate uncertainty throughout (via bootstrapping)
Copyright © 2022 Michael D. Garber